[PATCH v2 0/6] crash: Bundle of fixes for Xen

[PATCH v2 0/6] crash: Bundle of fixes for Xen

[Daniel Kiper]

Hi,

It looks that Xen support for crash have not been maintained
since 2009. I am trying to fix this. Here it is bundle of fixes:
  - xen: Always calculate max_cpus value,
  - xen: Read only crash notes for onlined CPUs,
  - x86/xen: Read variables from dynamically allocated per_cpu data,
  - xen: Get idle data from alternative source,
  - xen: Read data correctly from dynamically allocated console ring, too
    (fixed in this release),
  - xen: Add support for 3 level P2M tree
    (new patch in this release).

Daniel

Re: [PATCH v2 0/6] crash: Bundle of fixes for Xen

[Dave Anderson]

----- Original Message -----
> Hi,
> 
> It looks that Xen support for crash have not been maintained
> since 2009. I am trying to fix this. Here it is bundle of fixes:
>   - xen: Always calculate max_cpus value,
>   - xen: Read only crash notes for onlined CPUs,
>   - x86/xen: Read variables from dynamically allocated per_cpu data,
>   - xen: Get idle data from alternative source,
>   - xen: Read data correctly from dynamically allocated console ring, too
>     (fixed in this release),
>   - xen: Add support for 3 level P2M tree (new patch in this release).
> 
> Daniel
> 

Hi Daniel,

The original 5 updates specific to the Xen hypervisor look OK,
but new patch 6/6 is going to take some studying/testing to
alleviate my backwards-compatibility worries.  Can I ask whether
you fully tested it with older 2-level P2M tree kernels? 

Thanks,
  Dave

Re: [PATCH v2 0/6] crash: Bundle of fixes for Xen

[Daniel Kiper]

On Fri, Aug 10, 2012 at 03:11:57PM -0400, Dave Anderson wrote:
>
>
> ----- Original Message -----
> > Hi,
> >
> > It looks that Xen support for crash have not been maintained
> > since 2009. I am trying to fix this. Here it is bundle of fixes:
> >   - xen: Always calculate max_cpus value,
> >   - xen: Read only crash notes for onlined CPUs,
> >   - x86/xen: Read variables from dynamically allocated per_cpu data,
> >   - xen: Get idle data from alternative source,
> >   - xen: Read data correctly from dynamically allocated console ring, too
> >     (fixed in this release),
> >   - xen: Add support for 3 level P2M tree (new patch in this release).
> >
> > Daniel
>
> Hi Daniel,
>
> The original 5 updates specific to the Xen hypervisor look OK,
> but new patch 6/6 is going to take some studying/testing to
> alleviate my backwards-compatibility worries.  Can I ask whether
> you fully tested it with older 2-level P2M tree kernels?

As you asked me earlier I have tested all patches on Xen 3.1 and 4.1,
Linux Kernel Ver. 2.6.18 (P2M array), 2.6.36 (2 level P2M tree)
and 2.6.39 (3 level P2M tree). Additionaly, there were some internal
tests done by others in my company.

Daniel

Re: [PATCH v2 0/6] crash: Bundle of fixes for Xen

[Dave Anderson]

----- Original Message -----
> On Fri, Aug 10, 2012 at 03:11:57PM -0400, Dave Anderson wrote:
> >
> >
> > ----- Original Message -----
> > > Hi,
> > >
> > > It looks that Xen support for crash have not been maintained
> > > since 2009. I am trying to fix this. Here it is bundle of fixes:
> > >   - xen: Always calculate max_cpus value,
> > >   - xen: Read only crash notes for onlined CPUs,
> > >   - x86/xen: Read variables from dynamically allocated per_cpu
> > >   data,
> > >   - xen: Get idle data from alternative source,
> > >   - xen: Read data correctly from dynamically allocated console
> > >   ring, too
> > >     (fixed in this release),
> > >   - xen: Add support for 3 level P2M tree (new patch in this
> > >   release).
> > >
> > > Daniel
> >
> > Hi Daniel,
> >
> > The original 5 updates specific to the Xen hypervisor look OK,
> > but new patch 6/6 is going to take some studying/testing to
> > alleviate my backwards-compatibility worries.  Can I ask whether
> > you fully tested it with older 2-level P2M tree kernels?
> 
> As you asked me earlier I have tested all patches on Xen 3.1 and 4.1,
> Linux Kernel Ver. 2.6.18 (P2M array), 2.6.36 (2 level P2M tree)
> and 2.6.39 (3 level P2M tree). Additionaly, there were some internal
> tests done by others in my company.
> 
> Daniel

OK good.  It tests OK on a few older pvops kernels that I have on hand.

The only thing I've changed is to handle compiler warnings in x86_64.c and 
x86.c by initializing p2m_top to NULL in x86_64_pvops_xendump_p2m_l3_create()
and x86_pvops_xendump_p2m_l3_create().  I also used GETBUF() in those two
functions to avoid having to add the malloc-failure line.

Queued for crash-6.0.9.

Thanks,
  Dave

Re: [PATCH v2 0/6] crash: Bundle of fixes for Xen

[Daniel Kiper]

On Mon, Aug 13, 2012 at 03:16:53PM -0400, Dave Anderson wrote:

[...]

> OK good.  It tests OK on a few older pvops kernels that I have on hand.
>
> The only thing I've changed is to handle compiler warnings in x86_64.c and
> x86.c by initializing p2m_top to NULL in x86_64_pvops_xendump_p2m_l3_create()
> and x86_pvops_xendump_p2m_l3_create().  I also used GETBUF() in those two
> functions to avoid having to add the malloc-failure line.

No problem. However, I am a bit surprised that you have seen some warnings.
I have not seen any. Did you compiled crash with extra options or something
like that? Or maybe there is a diffrence in our compilers (mine
is gcc version 4.1.2 20061115 (prerelease) (Debian 4.1.1-21) - quite ancient).

> Queued for crash-6.0.9.

Thanks.

Daniel

Re: [PATCH v2 0/6] crash: Bundle of fixes for Xen

[Dave Anderson]

----- Original Message -----
> On Mon, Aug 13, 2012 at 03:16:53PM -0400, Dave Anderson wrote:
> 
> [...]
> 
> > OK good.  It tests OK on a few older pvops kernels that I have on
> > hand.
> >
> > The only thing I've changed is to handle compiler warnings in x86_64.c and
> > x86.c by initializing p2m_top to NULL in x86_64_pvops_xendump_p2m_l3_create()
> > and x86_pvops_xendump_p2m_l3_create().  I also used GETBUF() in those two
> > functions to avoid having to add the malloc-failure line.
> 
> No problem. However, I am a bit surprised that you have seen some warnings.
> I have not seen any. Did you compiled crash with extra options or something
> like that? Or maybe there is a diffrence in our compilers (mine is gcc version
>  4.1.2 20061115 (prerelease) (Debian 4.1.1-21) - quite ancient).

Extra options (also with gcc 4.1.2) -- try building with "make warn".  It adds:

 WARNING_OPTIONS=-Wall -O2 -Wstrict-prototypes -Wmissing-prototypes -fstack-protector

or "make Warn", which adds -Werror to the above.

In those two functions there are "goto err" statements prior to allocating
p2m_top, and so the free(p2m_top) could receive a random stack value.  Of
course even if it did get by the free() call in that highly unlikely case,
the crash session is just about to fatally shut itself down.

Thanks,
  Dave